Skip to Main Content
Until recently, most of WSD models have used adjacent words surrounding a target word as context information for word sense disambiguation. The difficulty of parsing Korean sentences and properly analyzing their structures restricted us to use syntactic relations to select proper senses of Korean words. In this paper, we propose the method to disambiguate Korean noun senses in unrestricted lexis using a statistical WSD model based on lexical and semantic information within syntactic relations. We disambiguate noun senses step by step in the range of local syntactic relations such as VP, NP and compound nouns. We classified the meaning of Korean nouns with about 400 semantic codes allowing for 8 levels in the semantic hierarchy and experimented for 1,838 homographs using the semantic information in consideration with syntactic relations. In the experiment, average senses of homographs that have different translations are 2.84 and the precision of word sense disambiguation is 86.2%.